Research map / Adversarial robustness and security / Black-box and transfer attacks
Hearing Without Noticing? Attention-Aware Stealthy Black-Box Adversarial Audio Attacks
Tianyi Xu, Cheng'an Wei, Yue Zhao, Kai Chen
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add masking-aligned carrier selection (ablation) | Ours · Physical attack ablation | Talking rate: 70.00% -> 10.37% | Improved |
| add masking-aligned carrier selection (ablation) | Ours · Physical attack ablation | SNR (dB): 14.81 -> 16.12 | Improved |
| add masking-aligned carrier selection (ablation) | Ours · Physical attack ablation | Normal rate: 7.85% -> 55.57% | Improved |
| add attention-dilution loss instead of magnitude constraint (ablation) | Ours · Physical attack ablation | Normal rate: 9.28% -> 55.57% | Improved |
| add attention-dilution loss instead of magnitude constraint (ablation) | Ours · Physical attack ablation | SNR (dB): 11.85 -> 16.12 | Improved |
| use Log Mel feature instead of MFCC (ablation) | Ours · Physical attack ablation | Normal rate: 35.73% -> 55.57% | Improved |
| use Log Mel feature instead of MFCC (ablation) | Ours · Physical attack ablation | SNR (dB): 9.84 -> 16.12 | Improved |
| use man with low voice target | Ours · Voice type (Whisper) | Normal rate: 55.57% -> 36.40% | Hurt |
| use woman with happy voice target | Ours · Voice type (Whisper) | Normal rate: 55.57% -> 77.30% | Improved |
| use human recordings as target | Ours · Voice type (Whisper) | Normal rate: 55.57% -> 50.00% | Hurt |
| Ours vs Kenku on stealthy samples under defenses | Ours · Defense: filtering | attack success rate: 7.69% -> 30.78% | Improved |
| Ours vs Kenku on perceptible samples under defenses | Ours · Defense: filtering | attack success rate: 61.54% -> 3.85% | Hurt |
Closest papers by meaning
- Adversarial Music: Real world Audio Adversary against Wake-word Detection System (NeurIPS 2019 · similarity 0.87)
- IO-RAE: Information-Obfuscation Reversible Adversarial Example for Audio Privacy Protection (AAAI 2026 · similarity 0.87)
- AdvWave: Stealthy Adversarial Jailbreak Attack against Large Audio-Language Models (ICLR 2025 · similarity 0.87)
- Adaptive Compression and Targeted Perturbation: A Unified Framework for Generalized Audio Deepfake Detection (NeurIPS 2026 · similarity 0.84)
- ALMGuard: Safety Shortcuts and Where to Find Them as Guardrails for Audio–Language Models (NeurIPS 2025 · similarity 0.83)
- Silence is Golden: Leveraging Adversarial Examples to Nullify Audio Control in LDM-based Talking-Head Generation (CVPR 2025 · similarity 0.83)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
